Suggestion Backlog Slog, 1/2/2017

[RyanCavanaugh]

• add enforceReadonlyAssignability flag for smoother transition to true readonly modifiers #13002 Add flag to enforce readonly in assignability relations
• Declare local variable as optional #13321 Allow ? in variable declarations
• Catch dependencies of functions used before assignment #13320 Detect invocations of functions which use uninitialized variables
• Checking for unsafe function overloads #13235 Verify correctness of function overload signatures
• Exhaustive checking of overloaded function return types #13225 Verify function overload correctness of return values
• Suggestion: throws clause and typed catch clause #13219 Allow functions to specify thrown exception types; infer catch variables from that
• Empty tuple type now meaningful (at least, in unions) #13126 Allow empty tuple types
• I sugguest outFiles, who can act M: N merge #13115 Multiple out files
• [Suggestion] Display the name of type alias in quick-info consistently #13095 Apply 'first type alias' rule locally in modules to simplify quickinfo
• Abstract class type argument inference #13087 Infer base class type parameters
• Array.join has improper typing in .d.ts file #13049 Disallow Array#join when T isn't string or number
• Type hole with compatibility between optional parameters/extra parameters #13043 Fix type hole around optional / ignored parameters
• Enums can not be used for index signature types #13042 Allow enums as index signature key types again
• (0,eval) produces TS2695 #12978 Allow the expression (0, eval)
• Type Inference in for... in combined with typeof #12944 Narrow x to ... & Function in typeof x === 'function'-guarded blocks
• Narrow "string" to "keyof T" with "in" guard? #12892 Narrow k to keyof typeof obj in k in obj-guarded blocks
• Reverse Enum Member Initialization Order for Constants #12693 Invert enum emit to improve minification when value is constant
• should warn/error on forward references during field initialization #12673 Disallow forward references in class member initialization
• [Suggestion] should Overloads not differ only by return type? #11281 Disallow trivially unreachable overloads

[RyanCavanaugh]

• add enforceReadonlyAssignability flag for smoother transition to true readonly modifiers #13002 Add flag to enforce readonly in assignability relations
  • Actually tracking at Interface with readonly property is assignable to interface with mutable property #13347
  • Wasn't added at first due to general flag avoidance
  • Is it simple to implement?
    • Somewhat
  • Breaks virtually every .d.ts file in the world
  • Would we allow mutable, or make it the default?
    • Default is mutable
  • Does this fork the world?
    • ... no?
  • Are there implications for literals?
    • No?
  • Probably as bad as strictNullChecks in terms of breakage but much less value gained
  • Revisit / AMF
• Declare local variable as optional #13321 Allow ? in variable declarations
  • Considered earlier
  • Potential for ES syntax conflict
  • No strong opinions either way
  • AMF
• Catch dependencies of functions used before assignment #13320 Detect invocations of functions which use uninitialized variables
  • Would have to follow all call graphs to correctly detect this in all cases
  • Too Complex
• Checking for unsafe function overloads #13235 Verify correctness of function overload signatures
  • Very expensive to calculate this (approximately worse than n^2)
  • Having more than one parameter complicates this immensely
  • Not clear anyone is messing this up?
  • Too Complex
• Exhaustive checking of overloaded function return types #13225 Verify function overload correctness of return values
  • Naive check isn't really good enough (union types should satisfy all constituent types' return values)
  • random check in comment applies against this argument instead of for it?
  • Too Complex
• Suggestion: throws clause and typed catch clause #13219 Allow functions to specify thrown exception types; infer catch variables from that
  • These exception types are never guaranteed
  • Too low value
• Empty tuple type now meaningful (at least, in unions) #13126 Allow empty tuple types
  • What's the index signature type of []?
    • never
  • Have to make sure an empty array is assignable to it
  • 👍 Accepting PRs
• I sugguest outFiles, who can act M: N merge #13115 Multiple out files
  • We would probably use this ourselves if it existed
  • But we recommend people do modules and bundling these days
  • See if other people want this since we hope --out is dying
  • AMF
• [Suggestion] Display the name of type alias in quick-info consistently #13095 Apply 'first type alias' rule locally in modules to simplify quickinfo
  • Works if the type alias is exported... kind of a bug
  • Wait for rewrite of the relevant code
  • Might be bugfixable
• Abstract class type argument inference #13087 Infer base class type parameters
  • Implications for Salsa
    • Salsa should probably instantiate base type parameters with any
  • How would we even instantiate A?
  • Generic JS classes should have T inferable from their constructors...
  • What about giving B the same type parameters as A?
  • What about inferring the type arguments from the super call?
  • What about transposing the type arguments down?
    • Assuming B had no type arguments, of course
  • Revisit for Salsa
    • A derived class with no type arguments and no ctor assumes its generic base class's type arguments
    • Otherwise we infer from the super call
• Array.join has improper typing in .d.ts file #13049 Disallow Array#join when T isn't string or number
  • 👎 we allow implicit string concat everywhere else

Remainder: Out of time

[zspitz]

@RyanCavanaugh What does AMF mean in this context?

[RyanCavanaugh]

Awaiting More Feedback